Enakta Labs - IT Press Tour #65 Dec 2025

Transcript

1. D E N I S N U J A - C E O IT Press Tour
2. About Enakta Labs Shipping Enakta Labs Platform v1.3 - Based on open source core DAOS 2.6.4 - Manages the full lifecycle of DAOS on bare metal deployments, with fully highly available management framework and backend - Fully featured high performance Microsoft™ SMB interface - High performance (limited API) S3 interface (tech preview) - PyTorch integration support - RoCE Support on 400Gbit ethernet (Thor2 and CX7, CX8 and Thor3 next) About Us - Startup founded in 2023 with focus on bringing DAOS based product to the wider enterprise market - Part of the original team behind the founding of the DAOS Foundation - Active contributor to the DAOS project - Part of the Nvidia Inception program
3. History of Enakta Labs - 2 founders : Denis Nuja & Denis Barakhtanov - Started working on DAOS in 2020 in our past companies, built another storage platform/system in the past together and delivered to a major government - 2023 we said 'enough' and founded Enakta Labs and became one of the original founders of the DAOS foundation (under Linux Foundation) - 2024 Enakta Labs Platform v1.0 started being tested with production workloads - We've been in the trenches longer than anyone else outside of the core DAOS team
4. Mission - We make the fastest storage engine on the earth usable and simple - No 400 page contracts - no dishonest roadmaps - no 'professional services' tax - If you have a support question at 3am, we pick up the phone. - We work for our customers
5. Pain and challenges we address - GPU idle waiting for data or checkpointing ? - 18-hour fsck ? - Having to over-provision your storage system to hit performance targets - Your developers or users want to use other ways of accessing their data ? - Need to become DAOS PhD to deploy and run it ? - You want to do something "weird" or "different" - You want to scale to supercomputer size systems ? We h a ve a solution for a ll of th a t ...
6. How does the platform look ?
7. We PXE boot the whole thing - Redundant (or single) management headnode (ISO boot and go) - Slim secure (SLES based) immutable in memory OS image - Storage and data services run in tight and light containers (podman) - We can deploy a few hundred node system in few hours if hardware and network is sorted (which almost never is ....) - HTTP/S UEFI PXE ... yes there is no TFTP server anywhere ...
12. Does it scale ?
13. IO 500 - Maximus-01 - Core42's 10,000 GPU cluster - Arista networking fabric - Thor2 400gbit ethernet NICs - (unfortunately) Gen4 NVMes - Converged and dedicated configuration
14. IO 500
17. Can we do better ? - Example of a single client performance we proved in field with RDMA on 400Gbit ethernet, writes are EC 16+1 (io500 config) - 3 different runs, starting to hit hardware limits a ccess bw(MiB/s) IOPS L a tency(s) block(KiB) xfer(KiB) open(s) wr/rd(s) close(s) tot a l(s) iter ------ --------- ---- ---------- ---------- --------- -------- -------- -------- -------- ---- write 278170 34772 0.003219 524288000 8192 0.004052 206.14 0.000117 206.15 0 re a d 254159 31770 0.003426 524288000 8192 0.004098 225.62 0.000089 225.62 0 + a ccess bw(MiB/s) IOPS L a tency(s) block(KiB) xfer(KiB) open(s) wr/rd(s) close(s) tot a l(s) iter ------ --------- ---- ---------- ---------- --------- -------- -------- -------- -------- ---- write 266490 33312 0.003326 524288000 8192 0.003935 215.18 0.000176 215.18 0 re a d 248892 31112 0.003538 524288000 8192 0.004391 230.39 0.000089 230.40 0 + a ccess bw(MiB/s) IOPS L a tency(s) block(KiB) xfer(KiB) open(s) wr/rd(s) close(s) tot a l(s) iter ------ --------- ---- ---------- ---------- --------- -------- -------- -------- -------- ---- write 275167 34397 0.003209 524288000 8192 0.004104 208.39 0.000098 208.40 0 re a d 255048 31882 0.003364 524288000 8192 0.003992 224.83 0.000086 224.84 0
18. Can we do better ? -We can -We scale out of box with new generation of NVMe drives and faster networks without code changes -If the NICs actually work correctly :)
19. The AMD question - Started validating AMD based deployments * ongoing work , unsupported at the moment * - Have not found a compelling argument for using AMD cpu's yet, only architectural choke points and problems
20. Use cases - HPC - AI/ML - Fintech - Media and entertainment Generally any customer where performance is primary or one of major metrics.
21. References - Beside the Core42 Maximus, no public references at this moment - Media and entertainment user will become public Q1/2026, system is used for volumetric capture and non linear editing using our SMB interface on our reference architecture done with Kioxia and Supermicro hardware (1PB usable) - Direct supplier to Google on R&D for Parallelstore (PyTorch integration) - Currently in final discussions and onboarding of 3 partners (2 in Europe, 1 in USA) and potentially one OEM partner (whitelabel)
22. Roadmap - S3 (v1.3 tech preview, v1.4 release) - NFS v4.2 (early internal testing, estimating Q2/2026 tech preview ) - Some discussions around tiering, but no commitment as of yet
 
 +
 
 Continuous improvement of the platform and hardware support, turn around on platform features is extremely short due to inherent design choices.
23. GTM - Single tier channel, partner first (we do not ship hardware and are hardware vendor neutral) - OEM+ discussion with large operator/s - Simple node based licensing with perpetual licenses - Quarterly or yearly subscription license for CSP/MSPs
24. Enakta Labs future - External funding discussions, but not interested in classic VC fueled marketing machine - Increase core DAOS development head count - Grow a fully capable 24/7 follow the sun support org And most importantly, even at a cost of faster/higher growth ... keep customers happy and work with them to build a product that actually makes a meaningful impact on their business.
25. 128 City Road, London, EC1V 2NX, United Kingdom
 https://www.enakta.com contact@enakta.com Enakta Labs